viewpoint_model_analysis.Rmd

---
title: "BSL Angle Processing analysis"
author: "Freya Watkins & Bodo Winter"
date: "2019-06-03"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

## Mixed model analysis:

Load packages:
```{r setup_load_packages}
setwd("~/R/Viewpoint")
library(tidyverse)
library(car)
library(lme4)
library(afex)
library(directlabels)
library(emmeans)
```

Read in the file, convert to tibble:
```{r read_in_file}
vp <- read.delim("2019-06-04 Viewpoint.txt")
vp <- as_tibble(vp)
head(vp)
vp$Angle <- as.factor(vp$Angle)
```

# Overview for methods section:

Check number of subjects and how many data points per subjects:
```{r check_subs}
length(unique(vp$Subject))
summary(vp$Subject)
```

Number of subjects per group; and then by L1 vs L2:
```{r check_groups}
table(vp$Group) / 490
```

What's the breakdown in terms of sign type?
```{r sign_types}
vpnoprac <- filter(vp, PrimeTarget != 'practice')
table(vpnoprac$SignType) / 90
```

What's the breakdown in terms of noun class?
```{r noun_classes}
table(vpnoprac$NounClass) / 90
```
Coded for noun class (body-anchored as well as other categories, following Pfau & Steinbach, 2005 - haven't found anything for BSL yet)

Check signs by degree of change: 
```{r sign_by_degree_change}
table(vp$Sign, vp$AngleChange)
```
Most items are balanced for degree of change. Some items are imbalanced due to the accidental inclusion of pairs with a 135 degree rotation (i.e. 90-45 on opposite sides) Other imbalances are due to different experiment versions not being perfectly counterbalanced across participants

Check sign by difficulty condition:
```{r sign_by_condition}
table(vp$Sign, vp$Difficulty)
```
Some items appear only in easier, only in harder, some appear in both. Only some appear in opposite, but all appear in same. This is due to the design aiming to balance for 'degree of change between prime/target' rather than 'Difficulty'.

Now we'll log-transform the RT variable for later analysis: 
```{r log_transform_RT}
vp <- mutate(vp, Log10RT = log10(RT))
```
The column "Priming" shows RT2-RT1; "LogDiff" shows LogRT2-LogRT1; "PercentFaster" shows how much faster RT2 was than RT1 as a percentage.

Number of unique signs:
```{r check_signs}
length(unique(vp$Sign))
```

Check number of data points per sign:
```{r check_signs_numbers}
table(vp$Sign)
```

Which signs don't have 90 data points per subject?

```{r lower_item_nums}
which(table(vp$Sign) != 90)
```
These are the 10 practice trials.

# Exclusions:

1. Discard practice trials
```{r no_practice}
vp <- filter(vp,
             PrimeTarget != 'practice')
```

2. Exclude items with low accuracy (below 70% overall)
Calculate accuracy per item:
```{r show_item_acc}
item_acc <- vp %>% group_by(Sign) %>% 
     summarize(ACC = mean(ACC))
print(item_acc)
```
13 items had below 70% accuracy (5.4% of 240 items)
Exclude those items from further analysis:
```{r no_low_acc_items}
vp <- filter(vp,
             ExclItem != 1)
```
Not sure about leaving this in or not? Perhaps a lower threshold than 70%? Some items have really low average acc e.g. 12%...

3. Exclude very fast (<500ms) or very slow (>10000ms) trials
```{r no_fast_slow_trials}
vp <- filter(vp,
             ExclSlowFast != 1)
```
10 trials out of 22,050 (0.05%) were excluded from further analysis.

# Creating tibbles
For the prime accuracy analysis, we should reduce the vp tibble to prime trials only:
```{r acc_prime_tibble}
vpAccPrimes <- filter(vp, PrimeTarget == 'prime')
summary(vpAccPrimes$PrimeTarget)
```
The prime accuracy analysis will be conducted on 10,800 trials.

For the prime RT analysis, we'll additionally exclude incorrect responses: 
```{r RT_prime_tibble}
vpRTPrimes <- filter(vpAccPrimes, ExclIncorrect != 1)
summary(vpRTPrimes$PrimeTarget)
```
The prime RT analysis will be conducted on 9685 trials

For the Accuracy difference analysis, we'll exclude prime-target pairs that do not neatly fit into the three difficulty categories. Firstly those pairs with a 135 degree rotation (e.g. 90L-45R; 54 pairs total = 0.53% of data); as well as pairs with a 'opposite' rotation (e.g. 45L-45R; 1383 pairs total = 13.54% of data) 
```{r acc_target_tibble}
vpAccTargets <- filter(vp, Excl135Deg != 1) %>% filter(ExclOpposite !=1) %>%  
                    filter(PrimeTarget == 'target')
vpAccTargets$Difficulty <- factor(vpAccTargets$Difficulty)
vpAccTargets$AngleChange <- factor(vpAccTargets$AngleChange)
summary(vpAccTargets$PrimeTarget)
```
The accuracy difference analysis will be conducted on 9362 prime-target pairs

For the RT difference analysis, we'll additionally exclude incorrect target trials (and their pairs), as well as the pairs of the 10 slow/fast trials. This is because we cannot calculate difference scores when one trial of a prime-target pair has been excluded, meaning its pair must be excluded too.
```{r RT_target_tibble}
vpRTTargets <- filter(vpAccTargets, ExclIncorrect != 1) %>% 
                  filter(ExclPairIncorr != 1) %>% filter(ExclPairSlow != 1)
vpRTTargets$Difficulty <- factor(vpRTTargets$Difficulty)
vpRTTargets$AngleChange <- factor(vpRTTargets$AngleChange)
summary(vpRTTargets$PrimeTarget)
```
The accuracy difference analysis will be conducted on 7987 prime-target pairs

## Descriptive statistics:

Averages per Group, Sign Type, Noun Class and Priming condition for accuracy:

```{r overall_averages}
vp %>% group_by(Group) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccPrimes %>% group_by(Group) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccTargets %>% group_by(Group) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccPrimes %>% group_by(SignType) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccTargets %>% group_by(SignType) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccPrimes %>% group_by(NounClass) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccTargets %>% group_by(NounClass) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vp %>% group_by(PrimeTarget, AngleLR) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccPrimes %>% group_by(AngleLR) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccTargets %>% group_by(Difficulty) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpAccPrimes %>% group_by(Subject) %>% 
  summarize(ErrorRate = mean(ErrorRate))
```
- Fluent signers clearly make fewer errors than intermediate M2L2 signers
- Prime/target breakdown: group differences more pronounced for prime trials; M2-L2 groups improve 2-3% for Targets. 
- 2HA signs least accurately comprehended. Biggest improvement at target for 2HS signs
- Midsagittal signs have fewer errors than body-anchored or lateral signs at Prime. Body-anchored signs still have high error rates at target
- Overall, slightly higher accuracy for target vs prime (learning effect?)
- A clear progression from 0 degree angle having the fewest errors (~6%), to 45 degree angles (~9%), to 90 degree angles having the most errors (~15%) for Prime trials
- For target trials, similar error rates across Difficulty conditions. 

Averages per Group, Sign Type, Noun Class and Priming condition for RT:
```{r overall_averages_RT}
vpRTPrimes %>% group_by(Group, AngleLR) %>% 
  summarize(RT = mean(RT))
vpRTTargets %>% group_by(Group) %>% 
  summarize(RT = mean(RT))
vpRTPrimes %>% group_by(SignType,AngleLR) %>% 
  summarize(RT = mean(RT))
vpRTTargets %>% group_by(SignType) %>% 
  summarize(RT = mean(RT))
vpRTPrimes %>% group_by(NounClass) %>% 
  summarize(RT = mean(RT))
vpRTTargets %>% group_by(NounClass) %>% 
  summarize(RT = mean(RT))
vp %>% filter(ExclIncorrect != 1) %>% group_by(PrimeTarget) %>% 
  summarize(RT = mean(RT))
vpRTPrimes %>% group_by(AngleLR) %>% 
  summarize(RT = mean(RT))
vpRTTargets %>% group_by(Difficulty) %>% 
  summarize(RT = mean(RT))
vpRTTargets %>% group_by(Difficulty, Group) %>% 
  summarize(Priming = mean(Priming))
vpRTTargets %>% group_by(Difficulty) %>% 
  summarize(PercentFaster = mean(PercentFaster))
```
- L2 intermediate signers about 480ms slower than L1 Deaf signers to Primes
- Gap between these groups closes to 365ms at target
- One-handed signs fastest, 2ha signs lag behind 2hs signs, same at Prime & Target
- No clear effect of noun class on RTs at prime or target
- Definite overall priming difference with targets around 150ms faster.
- 0 angle clearly fastest and both 90 angles slowest, but 45L much faster than 45R! (RH closest here?)
- Not obvious effect of difficulty condition on target RT alone
- But looking at Priming (RT2-RT1), obviously more priming in easier (229ms) & same conditions (194ms) than harder condition (71ms)
- Also clear from proportional transformation (% faster) 15% faster for easier, 14% for same, just 6% faster in harder condition.

Is there something going on between Sign Type and Angle?
```{r signtype_by_angle}
vpAccPrimes %>% group_by(SignType, AngleLR) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpRTPrimes %>% group_by(SignType, AngleLR) %>% 
  summarize(RT = mean(RT))
```
Less variation in error rates by angle to 1h signs. Error rate is lowest to 2hs at 0 and 45L. However when one hand may begin to occlude the other, particularly the right hand, error rate increases (i.e. at 90R for 2hs and 2ha). Interestingly 45L much better than 45R for 2hs signs.
For RT, again there is less variation cause by angle for 1H signs. 2ha signs are surprisingly slow at 0 - here 45L is the fastest. Clearest angle effect on 2hs signs - they start fastest of all at 0 but see huge slowdowns as angle increases likely due to occlusion.  45L faster than 45R in 1h and 2ha. 90R much slower than 90L for 2hs signs, i.e when the entire dominant right-hand is occluded.

What about between Noun Class and Angle?
```{r nounclass _by_angle}
vpAccPrimes %>% group_by(NounClass, AngleLR) %>% 
  summarize(ErrorRate = mean(ErrorRate))
vpRTPrimes %>% group_by(NounClass, AngleLR) %>% 
  summarize(RT = mean(RT))
```
Patterns for noun class are less clear. Largest variation in error rates for body-anchored signs and in RT for midsagittal signs. Some surprisng results - e.g. more errors and slower RTs for lateral signs at 90L where they should be clearest, than at 90R where they should be occluded.

## Plots

Accuracy plots:
```{r accuracy_plots}
vpAccPrimes %>% group_by(AngleLR, Group) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                         y = ErrorRate, group = Group, colour = Group)) +
  coord_cartesian(ylim = c(0,0.20)) +
  geom_dl(aes(x = AngleLR, y = ErrorRate, label = Group), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 18) +
  theme(legend.position = "none") +
  ggtitle('ErrorRate to Primes by Angle & Group') + xlab('Angle')
vpAccPrimes %>% group_by(AngleLR, L1L2) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                         y = ErrorRate, group = L1L2, colour = L1L2)) +
  coord_cartesian(ylim = c(0,0.20)) +
  geom_dl(aes(x = AngleLR, y = ErrorRate, label = L1L2), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 18) +
  theme(legend.position = "none") +
  ggtitle('ErrorRate to Primes by Angle & Group') + xlab('Angle')
vpAccPrimes %>% group_by(AngleLR, SignType) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                         y = ErrorRate, group = SignType, colour = SignType)) +
  coord_cartesian(ylim = c(0,0.15)) +
  geom_dl(aes(x = AngleLR, y = ErrorRate, label = SignType), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('ErrorRate to Primes by Angle & Sign Type') + xlab('Angle')
vpAccPrimes %>% filter(IconCat != 'na') %>% group_by(AngleLR, IconCat) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), y = ErrorRate, group = IconCat, colour = IconCat)) +
  coord_cartesian(ylim = c(0,0.15)) +
  geom_dl(aes(x = AngleLR, y = ErrorRate, label = IconCat), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('ErrorRate to Primes by Angle & Iconicity Category') + xlab('Iconicity Category')
vpAccPrimes %>% filter(IconRating != 0) %>% group_by(AngleLR, IconRating) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% ggplot +
  geom_point(aes(x=IconRating, y=ErrorRate, color=AngleLR, shape=AngleLR)) +
  geom_smooth(aes(x=IconRating, y=ErrorRate, color=AngleLR, shape=AngleLR), method=lm,se=FALSE) +coord_cartesian(ylim = c(0,0.20)) +
  ggtitle('ErrorRate by Angle & Iconicity Rating')
vpAccPrimes %>% group_by(AngleLR, NounClass) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), y = ErrorRate, group = NounClass, colour = NounClass)) +
  coord_cartesian(ylim = c(0,0.20)) +
  geom_dl(aes(x = AngleLR, y = ErrorRate, label = NounClass), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 18) +
  theme(legend.position = "none") +
  ggtitle('ErrorRate to Primes by Angle & Noun Class')+ xlab('Angle')
vpAccTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% group_by(Difficulty, Group) %>% 
  summarize(AccChange = mean(AccChange)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), y = AccChange, group = Group, colour = Group)) +
  coord_cartesian(ylim = c(-0.06,0.06)) +
  geom_dl(aes(x = Difficulty, y = AccChange, label = Group), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) + theme(legend.position = "none") +
  ggtitle('Change in Accuracy at Target by Difficulty & Group') + xlab('Difficulty')
vpAccTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% group_by(Difficulty, SignType) %>% 
  summarize(AccChange = mean(AccChange)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), y = AccChange, group = SignType, colour = SignType)) +
  coord_cartesian(ylim = c(-0.06,0.06)) +
  geom_dl(aes(x = Difficulty, y = AccChange, label = SignType), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) + theme(legend.position = "none") +
  ggtitle('Change in Accuracy at Target by Difficulty & Sign Type') + xlab('Difficulty')
vpAccTargets %>% filter(Difficulty != 'exclude', Difficulty != 'opposite', IconCat != 'na') %>% group_by(Difficulty, IconCat) %>% 
  summarize(AccChange = mean(AccChange)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), y = AccChange, group = IconCat, colour = IconCat)) +
  coord_cartesian(ylim = c(-0.06,0.06)) +
  geom_dl(aes(x = Difficulty, y = AccChange, label = IconCat), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) + theme(legend.position = "none") +
  ggtitle('Change in Accuracy at Target by Difficulty & Iconicity Category') + xlab('Difficulty')
vpAccTargets %>% filter(IconRating != 0, Difficulty != 'exclude', Difficulty != 'opposite') %>% group_by(Difficulty, IconRating) %>% summarize(AccChange = mean(AccChange)) %>% ggplot + geom_point(aes(x=IconRating, y=AccChange, color=Difficulty, shape=Difficulty)) + geom_smooth(aes(x=IconRating, y=AccChange, color=Difficulty, shape=Difficulty), method=lm,se=FALSE) +coord_cartesian(ylim = c(-0.06,0.06)) +
  ggtitle('Change in Accuracy at Target by Difficulty & Iconicity Rating')
vpAccTargets %>% filter(Difficulty != 'exclude', Difficulty != 'opposite') %>% group_by(Difficulty, NounClass) %>% summarize(AccChange = mean(AccChange)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), y = AccChange, group = NounClass, colour = NounClass)) +
  coord_cartesian(ylim = c(-0.06,0.06)) +
  geom_dl(aes(x = Difficulty, y = AccChange, label = NounClass), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) + theme(legend.position = "none") +
  ggtitle('Change in Accuracy at Target by Difficulty & Noun Class') + xlab('Difficulty')
```
Could only get IconRatings plot code to work as a scatter plot. Not sure if IconRating should be coded as ordered? Interesting that errors increase with iconicity for 90L (and slightly for 45L) whereas other angles show decrease in errors as iconicity increases. But not sure of the how strong these trends are... 

Plot RTs the same way:
```{r RT_plots}
vpRTPrimes %>% group_by(AngleLR, Group) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                         y = RT, group = Group, 
                         colour = Group)) +
  coord_cartesian(ylim = c(1500,2300)) +
  geom_dl(aes(x = AngleLR, y = RT, label = Group), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('RT to Primes by Angle & Group') + xlab('Angle')
vpRTPrimes %>% group_by(AngleLR, SignType) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                         y = RT, group = SignType, 
                         colour = SignType)) +
  coord_cartesian(ylim = c(1500,2300)) +
  geom_dl(aes(x = AngleLR, y = RT, label = SignType), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('RT to Primes by Angle & Sign Type') + xlab('Angle')
vpRTPrimes %>% filter(IconCat != 'na') %>% group_by(AngleLR, IconCat) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = 
                                      c("90L", "45L", "0", "45R", "90R")), 
                         y = RT, group = IconCat, colour = IconCat)) +
  coord_cartesian(ylim = c(1500,2300)) +
  geom_dl(aes(x = AngleLR, y = RT, label = IconCat), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('RT to Primes by Angle & Iconicity') + xlab('Iconicity Category')
vpRTPrimes %>% filter(IconRating != 0) %>%  group_by(AngleLR, IconRating) %>% 
  summarize(RT = mean(RT)) %>% ggplot +
  geom_point(aes(x=IconRating, y=RT, color=AngleLR, shape=AngleLR)) +
  geom_smooth(aes(x=IconRating, y=RT, color=AngleLR, shape=AngleLR), method=lm,se=FALSE) + coord_cartesian(ylim = c(1700,2100)) +
  ggtitle('RT to Primes by Angle & Iconicity Rating')
vpRTPrimes %>% group_by(AngleLR, NounClass) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot + geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                         y = RT, group = NounClass, 
                         colour = NounClass)) +
  coord_cartesian(ylim = c(1500,2300)) +
  geom_dl(aes(x = AngleLR, y = RT, label = NounClass), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('RT to Primes by Angle & Noun Class') + xlab('Angle')
vp %>% filter(ExclIncorrect != 1) %>% group_by(AngleLR, PrimeTarget) %>% 
  summarize(RT = mean(RT)) %>% ggplot + 
  geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")),
                         y = RT, group = PrimeTarget, 
                         colour = PrimeTarget)) +
  coord_cartesian(ylim = c(1700,2100)) +
  geom_dl(aes(x = AngleLR, y = RT, label = PrimeTarget), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 18) +
  theme(legend.position = "none") +
  ggtitle('RT to Primes/Targets by Angle') + xlab('Angle')
vpRTTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% group_by(Difficulty, Group) %>% 
  summarize(Priming = mean(Priming)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), 
                         y = Priming, group = Group, colour = Group)) +
  coord_cartesian(ylim = c(0,300)) +
    geom_dl(aes(x = Difficulty, y = Priming, label = Group), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('Priming at Target by Difficulty & Group') + xlab('Difficulty')
vpRTTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% group_by(Difficulty, SignType) %>% 
  summarize(Priming = mean(Priming)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), 
                         y = Priming, group = SignType, colour = SignType)) +
  coord_cartesian(ylim = c(0,300)) +
    geom_dl(aes(x = Difficulty, y = Priming, label = SignType), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('Priming at Target by Difficulty & SignType') + xlab('Difficulty')
vpRTTargets %>% filter(Difficulty != 'exclude', Difficulty != 'opposite', IconCat != 'na') %>%
  group_by(Difficulty, IconCat) %>% summarize(Priming = mean(Priming)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), y = Priming, group = IconCat, colour = IconCat)) +
  coord_cartesian(ylim = c(0,300)) +
    geom_dl(aes(x = Difficulty, y = Priming, label = IconCat), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('Priming at Target by Difficulty & Iconicity Category') + xlab('Difficulty')
vpRTTargets %>% filter(IconRating != 0, Difficulty != 'exclude', Difficulty != 'opposite') %>%  group_by(Difficulty, IconRating) %>% summarize(Priming = mean(Priming)) %>% ggplot + geom_point(aes(x=IconRating, y=Priming, color=Difficulty, shape=Difficulty)) +
  geom_smooth(aes(x=IconRating, y=Priming, color=Difficulty, shape=Difficulty), method=lm,se=FALSE) + coord_cartesian(ylim = c(0,300)) +
  ggtitle('Priming at Target by Difficulty & Iconicity Rating')
vpRTTargets %>% filter(Difficulty != 'exclude', Difficulty != 'opposite') %>%
  group_by(Difficulty, NounClass) %>% summarize(Priming = mean(Priming)) %>% 
  ggplot + geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), y = Priming, group = NounClass, colour = NounClass)) +
  coord_cartesian(ylim = c(0,300)) +
    geom_dl(aes(x = Difficulty, y = Priming, label = NounClass), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  theme_minimal(base_size = 16) +
  theme(legend.position = "none") +
  ggtitle('Priming at Target by Difficulty & Noun Class') + xlab('Difficulty')

```
Seems to be less going on with iconicity here, but interesting again that RT increase with iconicity for 45L only... Also not sure why low iconicity = more priming in Easier & Harder conditions only..?

Accuracy averages broken up by group with facet wraps:
```{r accuracy_group}
vpAccPrimes %>% group_by(AngleLR, Group) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")),
             y = ErrorRate, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) +
  coord_cartesian(ylim = c(0, 0.20)) +
  theme_minimal(base_size = 18) +
  ggtitle('Error Rate to Primes by Angle') + xlab('Angle')
vpAccPrimes %>% group_by(SignType, Group) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot(aes(x = SignType, y = ErrorRate, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) + 
  coord_cartesian(ylim = c(0, 0.15)) +
  theme_minimal(base_size = 22) +
  ggtitle('ErrorRate to Primes by Sign Type')
vpAccPrimes %>% filter(IconCat != 'na') %>% group_by(IconCat, Group) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot(aes(x = factor(IconCat, level = c("low", "mid", "high")), 
             y = ErrorRate, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) + 
  coord_cartesian(ylim = c(0, 0.15)) +
  theme_minimal(base_size = 22) +
  ggtitle('ErrorRate to Primes by Iconicity')+ xlab('Iconicity')
vpAccPrimes %>% group_by(NounClass  , Group) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot(aes(x = NounClass, y = ErrorRate, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) + 
  coord_cartesian(ylim = c(0, 0.15)) +
  theme_minimal(base_size = 18) +
  ggtitle('ErrorRate to Primes by Noun Class & Group')
vpAccTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% 
  group_by(Difficulty, Group) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
    ggplot(aes(factor(Difficulty, level = c("easier", "same", "harder")),
               y = ErrorRate, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) + 
  coord_cartesian(ylim = c(0, 0.15)) +
  theme_minimal(base_size = 18) +
  ggtitle('Error Rate to Targets by Difficulty') + xlab('Difficulty')
vpAccTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% 
  group_by(Difficulty, Group) %>% 
  summarize(AccChange = mean(AccChange)) %>% 
    ggplot(aes(x = factor(Difficulty, level = c("easier", "same", "harder")),
               y = AccChange, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) + 
  coord_cartesian(ylim = c(-.06, .06)) +
  theme_minimal(base_size = 18) +
  ggtitle('Change in Accuracy by Difficulty') + xlab('Difficulty')
```

RT averages broken up by group with facet wraps:
```{r RT_group}
vpRTPrimes %>% group_by(AngleLR, Group) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")),
             y = RT, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) +
  coord_cartesian(ylim = c(1500, 2300)) +
  theme_minimal(base_size = 18) +
  ggtitle('RT to Primes by Angle') + xlab('Angle')
vpRTPrimes %>% group_by(SignType, Group) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot(aes(x = SignType, y = RT, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) +
  coord_cartesian(ylim = c(1500, 2300)) +
  theme_minimal(base_size = 22) +
  ggtitle('RT to Primes by Sign Type')
vpRTPrimes %>% filter(IconCat != 'na') %>% group_by(IconCat, Group) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot(aes(x = factor(IconCat, level = c("low", "mid", "high")), 
                        y = RT, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) +
  coord_cartesian(ylim = c(1500, 2300)) +
  theme_minimal(base_size = 22) +
  ggtitle('RT to Primes by Iconicity')+ xlab('Iconicity')
vpRTPrimes %>% group_by(NounClass, Group) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot(aes(x = NounClass, y = RT, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) +
  coord_cartesian(ylim = c(1500, 2300)) +
  theme_minimal(base_size = 18) +
  ggtitle('RT to Primes by Noun Class')
vpRTTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% 
  group_by(Difficulty, Group) %>% 
  summarize(RT = mean(RT)) %>% 
    ggplot(aes(factor(Difficulty, level = c("easier", "same", "harder")), 
               y = RT, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) +
  coord_cartesian(ylim = c(1500, 2100)) +
  theme_minimal(base_size = 18) +
  ggtitle('RT to Targets by Difficulty') + xlab('Difficulty')
vpRTTargets %>% filter(Difficulty != 'exclude',
              Difficulty != 'opposite') %>% 
  group_by(Difficulty, Group) %>% 
  summarize(Priming = mean(Priming)) %>% 
    ggplot(aes(x = factor(Difficulty, level = c("easier", "same", "harder")),
               y = Priming, fill = Group)) +
  geom_bar(stat = 'identity', position = position_dodge()) + 
  coord_cartesian(ylim = c(0, 300)) +
  theme_minimal(base_size = 18) +
  ggtitle('Priming at Targets by Difficulty & Group') + xlab('Difficulty')
```

# By-subject plots
Create by-subject plots:
```{r subject_plots}
vpAccPrimes %>% group_by(AngleLR, Group, Subject) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")),
             y = ErrorRate, group=Subject, colour = Group)) +
  geom_point(size = 3, alpha = 0.5) + 
  geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                y = ErrorRate, group=Subject, colour = Group)) +
  theme_minimal(base_size = 16) +
  ggtitle('ErrorRate by Angle to Primes') + xlab('Angle')
vpAccPrimes %>% group_by(Angle, Group, Subject) %>% 
  summarize(ErrorRate = mean(ErrorRate)) %>% 
  ggplot(aes(x = Angle, y = ErrorRate, group=Subject, colour = Group)) +
  geom_point(size = 3, alpha = 0.5) + 
  geom_line(aes(x = Angle, y = ErrorRate, group=Subject, colour = Group)) + geom_dl(aes(x = Angle, y = ErrorRate, label = Subject), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  coord_cartesian(xlim = c(0, 100)) +
  theme_minimal(base_size = 16) +
  ggtitle('ErrorRate by Angle to Primes') + xlab('Angle')
vpRTPrimes %>% group_by(AngleLR, Group, Subject) %>% 
  summarize(RT = mean(RT)) %>% 
  ggplot(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")),
                        y = RT, group=Subject, colour = Group)) +
  geom_point(size = 3, alpha = 0.5) + 
  geom_line(aes(x = factor(AngleLR, level = c("90L", "45L", "0", "45R", "90R")), 
                y = RT, group=Subject, colour = Group)) +
  theme_minimal(base_size = 16) +
  ggtitle('RT by Angle to Primes') + xlab('Angle')
vpRTPrimes %>% group_by(Angle, Group, Subject) %>% 
  summarize(RT = mean(RT)) %>% ggplot(aes(x = Angle, y = RT, group=Subject, colour = Group)) + geom_point(size = 3, alpha = 0.5) + 
  geom_line(aes(x = Angle, y = RT, group=Subject, colour = Group)) + geom_dl(aes(x = Angle, y = RT, label = Subject), 
          method = list(dl.combine("last.points"), cex = 0.8)) + 
  coord_cartesian(xlim = c(0, 100)) + theme_minimal(base_size = 16) + ggtitle('RT by Angle to Primes') + xlab('Angle')
vpAccTargets %>% group_by(Difficulty, Group, Subject) %>% 
  summarize(AccChange = mean(AccChange)) %>% 
  ggplot(aes(x = factor(Difficulty, level = c("easier", "same", "harder")),
             y = AccChange, group=Subject, colour = Group)) +
  coord_cartesian(ylim = c(-0.2, 0.2)) +
  geom_point(size = 3, alpha = 0.5) + 
  geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), 
                y = AccChange, group=Subject, colour = Group)) +
  theme_minimal(base_size = 18) +
  ggtitle('Change in Accuracy by Angle (Acc1-Acc2)') + xlab('Difficulty')
vpRTTargets %>% group_by(Difficulty, Group, Subject) %>% 
  summarize(Priming = mean(Priming)) %>% 
  ggplot(aes(x = factor(Difficulty, level = c("easier", "same", "harder")),
             y = Priming, group=Subject, colour = Group)) +
  geom_point(size = 3, alpha = 0.5) + 
  geom_line(aes(x = factor(Difficulty, level = c("easier", "same", "harder")), 
                y = Priming, group=Subject, colour = Group)) +
  theme_minimal(base_size = 18) + 
  ggtitle('Priming by Angle (RT1-RT2') + xlab('Difficulty')
```
These are maybe too busy to use, but could be worth producing these by-subject graphs side-by-side for each group.

Leaving out the item-level analysis code for now (which will be more relevant to the other paper) - let's move on to the LMMs.

## Mixed Model Analysis, 1st half of experiment
First attempt at the mixed model for the Error analysis on Primes only (i.e. 1st half of the experiment)
```{r PrimeErrors_model}
PrimeErrors <- mixed(ErrorRate ~ Angle*Group  + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpAccPrimes, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(PrimeErrors)
summary(PrimeErrors)
```
Seems to converge... However I'm not sure if the random slope syntax is right (or for the following models). All participants see every sign once, but signs allowed to vary by angle (Angle). So diff ppts will see see same sign from diff angle. However, not all signs appear from every angle as items imbalanced.
Correlation of random effects probably too strong to be reliable - suggests something wrong with model? But if true, participants who have higher intercepts have steeper angle slopes. Makes sense that only the participants who found the task difficult at 0deg show a stronger interference effect of angle. (see Fig 81, p.271)

Multiple comparisons for the PrimeErrors model:
```{r primeAcc_model_pairwise}
PrimeAccPairwise = emmeans(PrimeErrors, specs = trt.vs.ctrl ~ Angle:Group, data = vpAccPrimes)
PrimeAccPairwise
```
Had to recode Angle as a factor in order to get it to work for multiple comparisons in emmeans
I specified '0,L1_Deaf' as the control (not sure if this is a good way to do it)

Now a first attempt at the mixed model for the RT analysis on Primes only (i.e. first half of the experiment)
```{r PrimeRT_model, message=FALSE}
PrimeRT <- mixed(Log10RT ~ Angle*Group + CorrResp + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpRTPrimes, all_fit = TRUE, method = "LRT", 
             control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(PrimeRT)
summary(PrimeRT)
```
This also seems to converge? But having trouble interpreting the results due to LogRTs.

Multiple comparisons for the PrimeRT model:
```{r primeRT_model_pairwise}
PrimeRTPairwise = emmeans(PrimeRT, specs = trt.vs.ctrl ~ Angle:Group, data = vpRTPrimes)
PrimeRTPairwise
```

And now I'll try checking the assumptions of these models:

## Checking assumptions
The below doesn't seem to be working - even though the models are now converging. Is the normality assumption checked differently for a mixed effect model? Think I got confused with sum-coding the fixed effects... but I did previously manage to to replicate the example in your book for a simple linear model based on my data.

{r check_assumptions}
Group <- rep(c('L1_Deaf','L1_Hearing','L2_Fluent','L2_Intermediate'), each = 3)
Angle <- c(0, 45, 90, 0, 45, 90, 0, 45, 90, 0, 45, 90)
Angle <- as.factor(Angle)
sum <- mutate(vpAccPrimes,
              Group_sum = factor(Group),
              Angle_sum = factor(Angle))
sum
contrasts(sum$Group_sum) <- contr.sum(4)
contrasts(sum$Angle_sum) <- contr.sum(3)

sum_mdl <- mixed(ACC ~ Angle_sum*Group_sum + CorrResp + (1 + Angle|Subject) +
               (1 + Angle|Sign), sum, method = "LRT", 
             control = lmerControl(optCtrl = list(maxfun = 1e6)))

res <- residuals(sum_mdl)
res

# Assess the normality assumption:
par(mfrow = c(1, 3))

# Plot 1, histogram:
hist(res) # distribution of residuals looks fairly normal
# Plot 2, Q-Q plot:
qqnorm(res)
qqline(res) # also indicates a good fit with normal distrib

# Plot 3, residual plot:
plot(fitted(sum_mdl), res) # all 12 cells have a fairly similar distribution
# possibly getting a bit larger for L2_Intermediate group further along y-axis

# check variance inflation factor:
vif(sum_mdl) # all btwn range of 1.0 - 1.05

## Mixed Model Analysis, 2nd half of experiment
Now we'll run LMMs on the data from the second half of the experiment. First modelling Change in Accuracy between Prime & Target as a function of Difficulty and Group.
```{r ChangeAcc_model}
ChangeAcc <- mixed(AccChange ~ Difficulty*Group + (1 + Difficulty|Subject), 
                   vpAccTargets, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(ChangeAcc)
summary(ChangeAcc)
```
```{r accchange_model_pairwise}
AccChangePairwise = emmeans(ChangeAcc, specs = trt.vs.ctrl ~ Difficulty:Group, ref = 3, data = vpAccTargets)
AccChangePairwise
```

Now modelling Priming (diff between logRT1 & logRT2) as a function of Difficulty & Group:
```{r Priming_model}
RTPriming <- mixed(LogDiff ~ Difficulty*Group + CorrResp + 
                     (1 + Difficulty|Subject), 
                   vpRTTargets, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(RTPriming)
summary(RTPriming)
```
```{r priming_model_pairwise}
PrimingPairwise = emmeans(RTPriming, specs = trt.vs.ctrl ~ Difficulty:Group, ref = 3, data = vpRTTargets)
PrimingPairwise
```

## Sub-analyses

# Sign Type Mixed Models
```{r SignTypeErrors_model}
SignTypeErrors <- mixed(ErrorRate ~ Angle*SignType + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpAccPrimes, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(SignTypeErrors)
summary(SignTypeErrors)
```
```{r Sign_type_errors_pairwise}
vpAccPrimes$Angle <- as.factor(vpAccPrimes$Angle)
STEPairwise = emmeans(SignTypeErrors, pairwise ~ Angle:SignType, data = vpAccPrimes)
STEPairwise
```


```{r sign_type_RT}
SignTypeRT <- mixed(Log10RT ~ Angle*SignType + CorrResp + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpRTPrimes, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(SignTypeRT)
summary(SignTypeRT)
```

```{r Sign_type_RT_pairwise}
STRTPairwise = emmeans(SignTypeRT, pairwise ~ Angle:SignType, data = vpRTPrimes)
STRTPairwise
```

```{r sign_type_AccChange}
SignTypeAC <- mixed(AccChange ~ Difficulty*SignType + (1 + Difficulty|Subject), 
                    vpAccTargets, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(SignTypeAC)
summary(SignTypeAC)
```

```{r Sign_type_AC_pairwise}
STACPairwise = emmeans(SignTypeAC, pairwise ~ Difficulty:SignType, data = vpAccTargets)
STACPairwise
```

```{r sign_type_Priming}
SignTypePriming <- mixed(LogDiff ~ Difficulty*SignType + CorrResp + (1 + Difficulty|Subject), 
                    vpRTTargets, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(SignTypePriming)
summary(SignTypePriming)
```

```{r Sign_type_Priming_pairwise}
STPPairwise = emmeans(SignTypePriming, pairwise ~ Difficulty:SignType, data = vpRTTargets)
STPPairwise
```

# Iconicity Mixed Models
```{r Iconicity_errors_model}
vpIconAccPrimes <- filter(vpAccPrimes, IconRating != 0)
IconicityErrors <- mixed(ErrorRate ~ Angle*IconRating + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpIconAccPrimes, all_fit = TRUE, method =
                 "LRT", control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(IconicityErrors)
summary(IconicityErrors)
```

```{r Iconicity_RT_model}
vpIconRTPrimes <- filter(vpRTPrimes, IconRating != 0)
IconicityRT <- mixed(Log10RT ~ Angle*IconRating + CorrResp + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpIconRTPrimes, all_fit = TRUE, method =
                 "LRT", control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(IconicityRT)
summary(IconicityRT)
```

`` `{r Iconicity_AC_model}
vpIconAccChange <- filter(vpAccTargets, IconRating != 0)
IconicityAC <- mixed(AccChange ~ Difficulty*IconRating + (1 + Difficulty|Subject), 
                     vpIconAccChange, all_fit = TRUE, method = "LRT", 
                     control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(IconicityAC)
summary(IconicityAC)
` ``
can't get this one to work: Error in anova.merMod(full_model, fits[[c]]) : 
  all models must be fit to the same data object

```{r Iconicity_priming_model}
vpIconPriming <- filter(vpRTTargets, IconRating != 0)
IconicityPriming <- mixed(LogDiff ~ Difficulty*IconRating + CorrResp +(1 + Difficulty|Subject), 
                          vpIconPriming, all_fit = TRUE, method =
                 "LRT", control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(IconicityPriming)
summary(IconicityPriming)
```

# Noun Class Mixed Models
` ``{r NounClass_errors_model}
NCErrors <- mixed(ErrorRate ~ Angle*NounClass + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpAccPrimes, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(NCErrors)
summary(NCErrors)
` ``

` ``{r NounClass_RT_model}
NCRT <- mixed(Log10RT ~ Angle*NounClass + CorrResp + (1 + Angle|Subject) +
               (1 + Angle|Sign), vpRTPrimes, all_fit = TRUE, method = "LRT", 
               control = lmerControl(optCtrl = list(maxfun = 1e6)))
print(NCRT)
summary(NCRT)
` ``

## Mental Rotation Analysis 

Read in Mental Rotation data file, convert to tibble:
```{r read_in_mr_file}
mr <- read.delim("2019-06-03 mental rotation data.txt")
mr <- as_tibble(mr)
head(mr)
```
Check number of subjects and how many data points per subjects:
```{r check_trials}
length(unique(mr$Subject))
summary(mr$Subject)
```
7 subjects have <48 data points (S108, S110, S125, S126, S142, S143, S144) due to technical problems (experiment file crashing).

Exclude very fast (<500ms) or very slow (>10000ms) trials:
```{r no_slow_trials}
mr <- filter(mr, ExclSlow != 1)
```

Summarise the mean Accuracy & RT by Subject for the mental rotation task:
```{r mr_subject_averages}
mrAcc <- mr %>% group_by(Group, Subject) %>% 
  summarize(MR_ACC = mean(ACC))
print(mrAcc)

mrRT <- filter(mr, ACC == 1)
mrRT <- mrRT %>% group_by(Group, Subject) %>% 
  summarize(MR_RT = mean(RT))
print(mrRT)
```
There are around 13/45 subjects with accuracy between 40-60% suggesting they may have been guessing/didn't fully understand the task. I excluded these the first time I did this analysis but have left them in for now.
Haven't log transformed RTs here - not sure if should?

Summarise the mean Accuracy & RT by Subject for the BSL experiment (prime data only): 
```{r bsl_subject_averages}
BSLAcc <- vpAccPrimes %>% group_by(Group, Subject) %>% 
  summarize(BSL_ACC = mean(ACC))
print(BSLAcc)
BSLRT <- vpRTPrimes %>% group_by(Group, Subject) %>% 
  summarize(BSL_RT = mean(RT)) 
print(BSLRT)
```
Bind tibbles:
```{r bind_tibbles}
AccCorr<- bind_cols(mrAcc, BSLAcc)
print(AccCorr)
RTCorr <- bind_cols(mrRT, BSLRT)
print(RTCorr)
```
This has duplicated the Group & Subject columns, I thought it would recognise they're the same and skip them? Oh well :)

```{r scatterplots}
ggplot(AccCorr, aes(x=MR_ACC, y=BSL_ACC, color=Group, shape=Group)) +
  geom_point() + 
  geom_smooth(method=lm,se=FALSE) +
  ggtitle('Mental Rotation Task Accuracy vs BSL Angle Task Accuracy')
ggplot(RTCorr, aes(x=MR_RT, y=BSL_RT, color=Group, shape=Group)) +
  geom_point() + 
  geom_smooth(method=lm,se=FALSE) +
  ggtitle('Mental Rotation Task RT vs BSL Angle Task RT')
```
  
Compute correlations between BSL task & Mental Rotation task for Accuracy & RT:
```{r correlations}
AccCorr %>% group_by(Group) %>%
  summarize(AccCOR=cor(MR_ACC,BSL_ACC))
RTCorr %>% group_by(Group) %>%
  summarize(RTCOR=cor(MR_RT,BSL_RT))
```
Not sure if this is the right way to calculate the correlations! Again these values seem quite high but maybe because R not Rsquared.

I've only looked at overall averages. In theory it would be possible to correlate by type of rotation e.g. 45deg sign vs 50deg shape rotation, 90deg sign vs 100deg shape rotation but I don't think there are enough items (only a very short test).
I wonder if it is worth reducing the data to just the conditions requiring rotation (ie. 45 & 90 for BSL; 50, 100, 150 for MR). We don't expect rotation to 0-degree signs, and there is no rotation required for two identical shapes.. So this correlation as is also has lots of non-rotation in the mix

In terms of interpreting the results, I would say that the RT correlation is of far more value than the Accuracy one, since the 3 fluent groups are performing at ceiling accuracy on the BSL task.

BSL task data per subject, 45 & 90 angles only:
```{r BSL_no0}
BSLAccNo0 <- filter(vpAccPrimes, Angle != "0")
BSLAccNo0 <- BSLAccNo0 %>% group_by(Group, Subject) %>% 
  summarize(BSL_ACC = mean(ACC))
print(BSLAccNo0)
BSLRTNo0 <- filter(vpRTPrimes, Angle != "0", ACC != 0)
BSLRTNo0 <- BSLRTNo0 %>% group_by(Group, Subject) %>% 
  summarize(BSL_RT = mean(RT))
print(BSLRTNo0)
```

MR task data per subject, 50, 100 & 150 rotations only:
```{r MR_no0}
mrAccNo0 <- filter(mr, Rotation != "0")
mrAccNo0 <- mrAccNo0 %>% group_by(Group, Subject) %>% 
  summarize(MR_ACC = mean(ACC))
print(mrAccNo0)
mrRTNo0 <- filter(mr, Rotation != "0", ACC == 1)
mrRTNo0 <- mrRTNo0 %>% group_by(Group, Subject) %>% 
  summarize(MR_RT = mean(RT))
print(mrRTNo0)
```

Combine tibbles for no 0 data:
```{r combine_no0_tibbles}
AccCorrNo0 <- bind_cols(mrAccNo0, BSLAccNo0)
print(AccCorrNo0)
RTCorrNo0 <- bind_cols(mrRTNo0, BSLRTNo0)
print(RTCorrNo0)
```

Create scatter plots for no 0 data:
```{r no0_scatterplots}
ggplot(AccCorrNo0, aes(x=MR_ACC, y=BSL_ACC, color=Group, shape=Group)) +
  geom_point() + 
  geom_smooth(method=lm,se=FALSE) +
  ggtitle('Mental Rotation Task Accuracy (no 0) vs BSL Angle Task Accuracy (no 0)')
ggplot(RTCorrNo0, aes(x=MR_RT, y=BSL_RT, color=Group, shape=Group)) +
  geom_point() + 
  geom_smooth(method=lm,se=FALSE) +
  ggtitle('Mental Rotation Task RT (no 0) vs BSL Angle Task RT (no 0)')
```

Compute no0 correlations between BSL & Mental Rotation tasks:
```{r no0_correlations}
AccCorrNo0 %>% group_by(Group) %>%
  summarize(AccCORno0=cor(MR_ACC,BSL_ACC))
RTCorrNo0 %>% group_by(Group) %>%
  summarize(RTCORno0=cor(MR_RT,BSL_RT))
```